Escolar Documentos
Profissional Documentos
Cultura Documentos
College of Materials Science and Engineering, Fuzhou University, University Park, 350108 Fuzhou, PR China
Quality and Reliability Laboratory, Institute of Materials Technology, Department of Mechanical Engineering, Universitt Kassel, Mnchbergstr. 3, 34125 Kassel, Germany
a r t i c l e
i n f o
Article history:
Received 18 September 2008
Received in revised form 2 February 2009
Accepted 7 February 2009
Available online 20 February 2009
Keywords:
Reliability assessment
Lifetime distribution
Finite element analysis
Probabilistic fracture mechanics
Fatigue crack growth
a b s t r a c t
Reliability assessment is an essential step to promote advanced materials and components into applications. In this paper, a general reliability assessment framework was proposed to predict the lifetime distribution of a structural steel component with inherent aws. By combining materials information, nite
element analysis of the stress eld and probabilistic fracture mechanics model, the distribution of failure
probability subjected to fatigue loads was predicted. The local failure probability distributions identify
the critical regions of the component visually. Both the global failure probability and the local failure
probability distribution can be considered as essential and fundamental data in structure design and system maintenance. Focus was placed on the probabilistic fracture mechanics model and fatigue crack
growth model.
2009 Elsevier Ltd. All rights reserved.
1. Introduction
Structural safety and reliability assessment is essential work for
design and maintenance of modern engineering systems. The systems range from microelectronic and bio-medical devices to large
machinery and structures, as well as civil engineering [14]. For
example, in a commercial project concerning a high efcient gas
turbine rotor in European Union, the full project was subdivided
into ve work packages, and one of them was to develop a reliability assessment tool for the steel component made from a nano-precipitate hardened high nitrogen steel. The reliability assessment
tool integrates material properties, nite element stress analysis
and probabilistic fracture mechanics consideration. The probabilistic fracture mechanics model is generally based on the assumption
that failure occurs due to the subcritical and catastrophic crack
growth of crack-like defects introduced during fabrication. Such
defects are initially present with a given probability, and are found
during pre- and in-service inspections with a probability depending on their size. The subcritical and catastrophic growth of these
defects is governed by fracture mechanics considerations, which
may also involve material properties that are randomly distributed. Cracks found by inspection determine the component to be
retiring or not.
The basic ideas of the reliability assessment tool are presented
in this paper. The probabilistic fracture mechanics model is described in Section 2. In Section 3, a semi-elliptical surface crack
growth model subjected to fatigue load is discussed and programmed. The demonstration of the reliability assessment tool is
given in Section 4 with a four-point-bending bar. A summary and
outlook are presented in Section 5.
1883
Nomenclature
a
Px
X
Compared with the volume aws, the surface aws are believed
to be more dangerous in metallic materials, so the surface cracks
are considered as the risk inherent aws in this work.
Probabilistic fracture mechanics [57] is used to describe the
failure behaviour induced by inherent cracks. Unstable propagation, i.e. spontaneous failure, occurs if the crack size exceeds a certain critical value ac. The random orientation of a crack in the
applied stress eld r in general leads to a mixed-mode load of
1884
the crack with the mode I, II, III stress intensity factors KI, KII and
KIII, respectively.
p
a YI
p
K II sII a Y II
p
K III sIII a Y III
K I rn
1
2
3
where YI, YII and YIII denote the corresponding geometric correction
factors, rn, sII and sIII denote the projections of the stress tensor on
the crack plane. Obviously, the values obtained for the projections
of the stress tensor depend on the orientation x of the crack plane.
The failure of cracks subjected to the mixed-mode load is described in terms of the failure criterion g(KI, KII, KIII). Failure occurs
if g exceeds a critical value gc
system, and 2p is the normalization factor of the uniform distribution of the random orientations of cracks normal to the surface.
The failure probability of a component containing exactly one
crack with random size, location, and orientation failing because
of the unstable propagation of the crack is obtained by multiplying
the probability P F a , Px, and Px and summing over all possible locations and orientations of the crack
1
PF
1
A
Z
A
1
2p
Z Z
ac x;x
fa0 adadXdA
12
PS 1 PF
13
n
PS
gK Ieq ; 0; 0 gK I ; K II ; K III
PS 1 PF n
g gc
K Ieq K Ic
p
K Ieq req Y I a
req depends on the local value of the stress tensor r and on the orientation x. Eq. (7) is only valid if req can be considered constant
along the crack size a, i.e. for small stress gradients. In the source
code, different criteria for the determination of the equivalent stress
req are prepared for choice by user. From Eq. (7), the critical crack
size ac is determined by
ac
K IC
2
PF a
14
Pn
M n eM
n!
15
The survival probability PS,A of a component containing an arbitrary number of surface cracks is obtained by multiplying the probn
ability Pn with the corresponding survival probability P S and
summing over all possible numbers of cracks
1
X
Pn PS
16
n0
Taking into consideration the above two Eqs. (14) and (15) and
the series expansion rule for the exponential function, Eq. (16)
leads to
1
PS;A expM PF
17
1
ac x;x
PS;A
8
Y rmax
fa0 ada
1
dA
A
1
Px fX xdX
dX
2p
Px fA xdA
10
11
PF;A 1 expM PF
18
M A M0
19
PF;A 1 expM 0 A PF
20
1885
0.8
PF
Crack Size with LogNormal Distribution
0.6
f(a)
0.4
Failure domain
0.2
PFa =
0.0
ac
0.00
0.10
0.20
0.30
0.40
0.50
Crack size, a (mm)
0.60
0.70
PF a
ac;0 x;x
fa0 ada
21
1
A
1
2p
Z Z
1
ac;0 x;x
fa0 adadXdA
22
where a(c,0)(x, x) is the critical initial crack size, which will reach the
critical size ac after N cycles loading. a(c,0)(x, x) depends on the
location of the crack, its orientation, material properties and load
conditions. Predening the cycle number of N, a(c,0)(x, x) can be
determined by integrating the crack growth law and inserting the
result into the failure criterion, see Section 3. An efcient numerical
algorithm for the solution of this problem has been nished. The
transformation of the failure domain simplies the question very
much since it is very difcult to obtain the probability density
function of the crack once the subcritical cracks start to grow under
fatigue loading.
Considering all the random variables [8] involved in the
fatigue process, the failure probability that a specic crack starts
to extend in an unstable manner described in Eq. (22) can be
treated as
1
PF
fz zdz
23
failure domain
where z is the vector of all random variables. There are three types
of random variables
(1) Random variables with limited amount of scatter with comparatively good database. These are the parameters of the
crack growth law (C, m), the fracture toughness (KIc), the
aspect ratio of the cracks (a0/c0) at the beginning of the lifetime, the applied loads (rmax, rmin) and the component
geometry. In a probabilistic analysis, these variables can be
dealt with in a crude Monte Carlo simulation.
(2) Random variables with broad scatter, good database and
strong inuence on the results. These are the micro-cracks
location x and orientation x. In this study, only homogeneous and isotropic materials are considered. This implies
that uniform distributions have to be used to describe the
scatter of the location x and of the orientation x of the
cracks. An efcient integration algorithm is needed in order
to obtain good estimates of the failure probability. In this
work, the Gauss quadrature rule for each nite element is
employed.
(3) Random variables on which very limited information is
available. This is the crack size distribution at the beginning
of the lifetime (fa0 a). A library of statistical models is provided in the package from which the user could select a suitable one.
Keeping these considerations in mind, the failure probability
can be written as
0.8
1
PF
Z
range of Z 1 ;...Z k
0.6
f(a)
ac
Z
range of Z 1 ;...Z k
0.00
0.10
0.20
Z pZ
0
24
where Z1, . . ., Zk are random variables of type (1). Solving the innermost integral of Eq. (24) leads to
QF
a(c,0)
1 1
A p
Failure domain
0.2
0.0
failure domain
da
= C( K A )m
dN
0.4
fZ1 z1 . . . fZk zk
0.30
0.40
0.50
Crack size, a (mm)
0.60
Fig. 3. The schematic failure domain during the cyclic load process.
0.70
fZ1 z1 . . . fZk zk
1 1
A p
Z p
0
25
1886
da
CDK A m
dN
dc
CDK C m
dN
26
27
where C and m are the material parameters during the fatigue process. c and a are the major and minor semi-axis of the semi-elliptical crack, respectively. DKA and DKC represent the stress intensity
factor ranges in deepest point A and on surface point C, respectively.
The stress intensity factor range DK involves the geometry factors. Based on NewmanRaju Equation [10], Fett [11] proposed an
extension geometry factor formula for the semi-elliptical surface
cracks, which covers the range 0 6 a/c 6 1.8. The general geometry
factor is given in Eq. (28)
r!
a10
a
Y I 1:13 0:1
tanh0:1
c
c
p h
2
14
i
p
a
2
2
cos2 h sin h
1 0:1 1 sin h
c
Ek
maxK A ;K C K Ic
da
CDK A m C
dN
m
a112 Dr p
pa
1:13 0:00997
c
Ek
30
33
34
35
36
a an jNNn
37
When solving the system of equations containing ordinary differential equation, generally, we separate the variables and convert to its
integration form
Nn
an
a0
an
1
DK A m da
C
m
a112 Dr p
1
pa
da
1:13 0:00997
C
c
Ek
38
Once the initial conditions K Ic ; rmax ; rmin ; C; m; a0 ; a0 =c0 are given, solving Eqs. (29)(35), we can get the critical crack size ac under the failure condition. Thereafter, we integrate the crack growth
Eq. (38) from a0 to an (an = ac) and get the fatigue cycles to failure.
Due to the complicacy of the system of equations, the numerical
solution is employed. The relationship between the load cycles to
failure and the initial conditions can be established numerically
as follow
32
a a0 jN0
a0
28
31
a112 r p
pa
K A 1:13 0:00997
c
Ek
a112 r p
a12
pa
K C 1:13 0:00997
1:1
c
Ek
c
2
"
!#m2
m2
c m2
2
a
a0 2
0
m
m
1 1:1
1:1
c
a
c0
1
a1:65 2
Ek 1 1:464
; when 0 < a=c 1
c
a1:65 12
; when 1 < a=c < 1:8
Ek 1 1:464
c
39
Lifetime, Cycle
1E6
1E5
1E4
KIc=50 MPa(m)
c=1.36E-10
m=2.25
0.5
1E3
1E-3
Fig. 4. The semi-elliptical surface crack in a plate with thickness t and width w.
0.01
0.1
Initial crack size, a (mm)
Fig. 5. The schematic relationship between the initial crack size and lifetime.
1887
Table 3
The reliability assessment for the four-point-bending bar under cyclic load.
Load case
Load value
rmax ; MPa
rmin MPa
250
50
250
50
40,000
Fig. 8
0.0064
0.99352
50,000
Fig. 9
0.38045
0.61955
Table 1
The parameters of the initial cracks.
Aspect ratio
(a0/c0)
Crack size
Distribution
type
Mean
(mm)
Variance
0.75
50
Lognormal
distribution
0.10
0.0016
Table 2
The fatigue crack growth parameter of material (mean).
p
KIc (MPa m)
C
50
10
1.3610
2.25
Fig. 7. The FEM model, displacement and Von Mises stress distribution of the fourpoint-bending bar in the nite element analysis under unit load. The type of nite
element is C3D8.
1888